Crafting the Perfect Model: Fine-Tuning and Merging LLMs

Join Maxime Labonne, Staff ML Scientist at Liquid AI, as he dives into the world of fine-tuning and merging large language models (LLMs). This presentation covers:

• Scenarios where fine-tuning is most beneficial.

• Key methodologies for supervised fine-tuning, including LoRA and QLoRA.

• Preference alignment methods like DPO.

• Model merging techniques such as SLERP, DARE, passthrough, and frankenMoEs.